PAUL J. HOPPER, State University of New York at Binghamton

GLOTTALIZED AND MURMURED OCCLUSIVES lN INDO­EUROPF.AN'

/ndo-Europeanists have traditionally reconstructed for Proto­Indo-European (PIE) a system of stops consisting of two voiced members (/d/ and /dh/) and either one or two voiceless (, t/ or /tl and /th/). For those who posit a three-stop system /t/, , d1, and /dh/, the problem of typological improbability presents itself, for a typologically plausible triple stop system should have only one voiced stop. In this paper, two main points are made: (1) the voiced aspirates should be regarded as partially voiced (murmured) rather than as aspirated or tense, and (2) the supposed 'plain voiced' stops ( /d/, etc.) show many of the typological charac­teristics of glottalized stops (ejectives), e.g. they are excluded from "inflectional affixes, they may not cooccur with one another in the same root, etc. Combining these two observations, we may state the constraint in PIE which prevented roots of the type *tebh in a highly plausible way: two nonglottalized occfusives must agree in voicing. This formulation is possible because the murmured stops are the only ones with voicing. Areal and typo­logical consequences of this hypothesis are mentioned, and a possible trajectory of tht; proposed 1 t/ ./ t' / --'d / sy.stem in Ger­manic and other IE dialect groups is discussed.

1. In the enterprise of reconstructing proto-1 anguages, I i ngui sts have customarily assumed that the hypothetical entities which are posited are constructed within the same parameters as those found in extant languages, It is only fairly recently, however, beginning perhaps with Jakobsen's paper Typological Studies and their Contribution to Historical Comparative Linguistics (1957), that the value of language typologies and language uni­versals has been recognized as a working method.

HOPPER: IE Occlusives- 141

Typological studies may be used in at least two ways, as a control and as a discovery procedure. In the first of these uses, a linguistic system which may seem to be realistic on purely comparative grounds may be either confirmed or rejected when the typological evidence is examined. As on exam pie, we may take Dempwolff's reconstruction of the Prato-Austronesian con­sonant system, Dempwolff (1934) posited stops in voiced and voiceless pairs at five points of articulation, and a single frica­tive, the voiced velar fricative 'r ', A system of this kind would have to be viewed as unnatural to a degree where its existence could confidently be denied.

The use of typological studies as a discovery procedure in comparative li ngui s tics is seen in the application of i mpl i cationa I universals, i.e. statements of the kind: 'If a language has feature X, it wi II also have feature Y .' As a m otter of fact, c omparati vi sts have probably always worked implicitly with projections of this kind. But it is only since universals hove been stated explicitly that linguists have consciously applied them in recovering de­toils of the structure of proto-languages. From the widespread presence in earlier Indo-European languages of a comparative adjective construction of the type Latin melle dulcius 'sweeter than honey', with the word order standard noun plus comparative, and one or two other diagnostic features, we may ·infer that Proto­Indo-European belonged to a type of language in which the direct object of the verb preceded the verb, where there was a tendency for modifying elements to precede the modified element (genitive plus noun rather than noun plus genitive, etc.), and so· on; cf. Lehmann 1972, Greenberg 1966. .

2. A Typological Critique of the Indo-European Obstruent System as Usually Reconstructed

In this paper the generally assumed obstruent system of Proto-Indo-European will be examined from the point of view of phonetic and systematic phonemic universals. I shall be con­cerned principally with the manner of articulation of the stops.

During the present century, the main changes which have come about in notions about the Proto-Indo-European consonan.t system have been the result of (1) the positing of further conson­ant phonemes through the laryngeal theory, and (2) the re-evalu­ation of systematic aspects cf Indo-European phonology through theoretical advances, mainly in phonological theory. The phonetic nature of the system of stops which the Neo-Grammori ans recon­structed was rorel y discussed.

gios>"o 7·2 11973:

However, several writers hove at different times expressed reservations about the phonetic realism of the system propounded by Brugmonn, whiie continuing to live with systematic aspects of the scheme. For example, Prokosch I 1938: 39-41), while re­jecting !he ide a of a voiced osp1 rote series, continues to use the symbols bh, dh, etc., portly for the sake of conformity to other works, and portly because his interpretation of these sounds as Ienis unvoiced fricatives does not affect his exposition of the facts of !he obstruent changes as they occurred in Germanic. Similarly, H1rt 11927: 218-219) is able to get by with a few brief paragraphs of discussion of the voiced aspirates, and after point­ing out that a 'true' vo1ced asp1rate such as the segment sequence in the middle of the word Rebhuhn pronounced as a sing! e segment is practically impossible from a phonetic point of view, he con­tinues throughout the work to represent the voiced aspirates as units composed of voiced occlusives plus h.

Other writers are equally agnostic about the phonetic pro­perties of the reconstructed units of sound. Meillet (1936: 88) says of the voiced aspirates that they were characterized by 'a glottal articulation which we have no means of determining exact! y'. And according to Lehmann ( 1955: 11): 'From a structural pcint of vi "!W the phonetic interpretation is of secondary importance.' Discuss ion of the phonetic realization of the phonological units of Indo-European has, in fact, occupied a rather marginal place in comparative studies of the language family in the twentieth century. Whatmough's suggestion 11937) that the labiovelors were characterized phonetically by coarticulation ([kp], etc.), a type of phoneme common in African I anguages of the West Atlantic

Coast, appears 1o have giv.en rise 1o no dis cuss ion among other

I ingui sts. • 2.2 Few comparative linguists these days would be content with a pure! y algebraic account of sound changes. In recent years, linguists have insisted with increasing emphasis that language change and hypothetical proto-languages should not only have internal consistency, but should also conform to some standard of 'naturalness', i.e. should have generalizable analogs in docu­

mented systems and diasystems. It is clear that the task of the historical linguist must now be seen as that of defining diachronic naturalness and finding empirical evidence for what constitutes natural as opposed 1o unnatural systems. There remains in addi­tion the awesome task of devising an adequate formalism for ex­

pressing hypothesized natural systems and changes, many of which are assumed 1o be natural solely because of their cross­( i ngui sti c frequency of occurrence." The present article is intended

as a serious attempt 10 discuss 1he obstruent system of Proto­Indo- European in the light of 'natural' phonological systems. Accordingly it should be said in advance that the attempt wilt suffer from the inevitable defects of inadequate formalization and a certain recourse 1o impressionism with respect to 'plausi­

bility'.

2.3 _The Generally Assumed PIE Obstruent System. The Proto-l.ndo-European obstruent system generally assumed by scholars is sul1stanti ally that set forth in Lehmann's Prato- indo-European

Phonology (1955). This system assumes three points of articula­tion: labial, dental, and velar (excluding laryngeals); three glat­talic types: unvoiced, plain voiced, and voiced aspirate; and a contrast in the velar series between labialized and unlabialized. The system also has a single fricative, lsi (again excluding iaryngeals), and displays a surprising gap in the labial series, the plain voiced stop ""/b/ being absent. Figure T shows the acclusivPo; as phonemic units, and Figure 2 d1splays the same

LABIAL DENTAL VELAR

VOICELESS I ' / pl /tl /ki, lk '/

VOICED /d/ fg/, lgOf

VOICED ASPIRATE /bh/ 1dh/ 'gh/, !q- hI

Figure 1.

p ""b bh t d dh k g gh k~ g'·' gch

obstruent + - '-

continuant

anterior

coronal - - '- ( --) (_) (-) (-) (-) (-)

round (-) (-) (-} (-) (-) (-}

voiced

tense 1--} !-) -- ( --) - ( .)

Figure 2

9;0SS'J 7.2 '197J'

information as a distinctive feature matrix; the articulatory features used llre those of Chomsky and Halle (1968). The use of the feature 'tense' to differentiate plain and aspirated voiced stops has be­come customary in eli scussions of Indo-European phonology, but

it will be examined critically in a later section. For the purposes of the present paper, l shall assume that the points of articula­tion are adequate! y specified, although it should be pointed out that the r.on-labialized velars were possibly palatalized; since they develop into palatals in many dialects (satem languages). Tenseness is redundant in the labial series if .. /b' is not assumed, for /bh/ is set off &om /p/ by voicing alone. -

2.4.0 Discussion of Phonetic Features of PIE Stops.

2.4.1 The Two Voiced Stops. As has often been noticed(most recently by Peeters, 1971), at each point of articulation except the labial there is a triple contrast, necessitating two distinctive features with a redundancy in one specification, viz. either:·

d dh

voiced

tense 0

or

tense

voiced 0

It is now possible to argue that the VOICed aspirates may have had voiceless allophones or dialectal variants, since the presence or absence of voicing would not affect the triple contrast. The appearance of voiceless sounds in Greek and Italic is thus readily explained. The proposals which I shall make in this paper do not, as far as I can see, ultimately conflict with this suggestion.

2.4.2 Aspiration and Tenseness. It has become customary in recent phonological studies of Indo-European and of the Indo­Aryan languages to assign the feature l+ tense] to the presumed voiced aspirates. The main advantage of this specification lies in the fact that it permits four manners of articulation to be ac­counted for with two features:

HOPPER: IE Occlusives -145

t th d dh

-voic.ed- + +

tense - + - +

The justific.ation for tenseness disappears, however, when deal­ing with a language having no voiceless aspirates. We must ask the question whether lenseness in physiological-phonetic terms

___ con be assumed as a func.tional articulatory foetor in Proto-Indo­European, and. if so, whether tenseness would be realized as a .;pe- of phoneme characterized by voicing and aspiration.

2.4..2.1 The second of these two d aims, that the third serie-s of obstruents was wiced cr~d ospi rated, has never been held with great enthusiasm by lndo-Europeani sts. None of the attempts to replace the voiced aspirate hypothesis has found wide accep-tance, ho-ver. .

Probably the main reason why lpclo-Europeanists continued to use the ~voiced aspirate' convention has been the implicit priority of Sanskrit phonology. It is true thai the primacy of Sans­krit had been repeatedly disproven through such discoveries as the original vowel system in e/o, the syllabic nasals, and the apparently derivative nature of the voiceless aspirates. Nonethe­less, the Sanskrit 'voiced as pi rates' caul d be shown to represent a doss of sounds which sho-d up as fricatives in Italic and Germanic, and which furthermore patterned in important respects like the 110iceless aspirate stops, which were phonetically estab­lished for Sanskrit. Thus Grassmann's law operated not only on forms containing two voiced aspirates inherited from the proto-1 anguage, but also on derivative roots containing two vo i eel ess aspirates (lchya- 'see', cakhyau (pf.)). The nsle could be stated neat! y as an across-the-board di ssimi I otion of aspirates, voiced or unvoiced. Sanskrit also showed a morphophonemic alternation of voiced aspirates with a phoneme transcribed as /h ', as in hanti lghnanti 'he strikes/~ey strike', represent! ng a Proto-Indo­European ablaut alternation 'g'·hen-/g'"hn-.

The correspondences in other Indo-European languages were also strongly suggestive of the analysis of _the third series as voiced stops 'pi us' something else. That this ~xtra feature should be as pi ration accounted for the Greek ref! exes ph, th, and k h, which were known ro be 110iceless aspirates, and made the de>~elopment into fricatives under some circumstances in Italic and Germanic plausible.

tf the term 'aspiration' as used in conjuncttot' with the des-

glossa 7:2 (1973)

cription '¥CHced stop' is taken literally, then phonetic observa­tion would have b IUie aut completely the possibility. of such a combination-of features. An aspirated obstruent is one character· i zed by a pause between the offset of the obstruent and the i nit­iation of voicing in the- following vowel (or the articulatory motion of obstruency in the following consonant). During this pause, phonation continues, with the vocal organs generally in the articulatory position of· the following phone. With voiceless aspirates this ~imply means that voicing in a following vowel is delayed. In the case of the supposed wiced aspirates, however, voicing waul d have to be momentari I y suspended between· the voiced obstruent and the voicing of the vowel. Such a suspension, while !10t physiologically impossible, is inconvenient and, lin­guistically, highly improbable. The two components of the med i al cluster in the German word Rebhuhn (Hirt's example) are both unvoiced, and are in any case separated by a very distinct phono­logical boundary. The situation is summed up by Ladefoged (1971: 9), who, after describing hypothetical 'voiced aspirates' in these terms, remarks that 'such a sound has not yet been ob­served in any language.'

2.4.2.2 The theory that one of the distinctive features mark­ing the· PIE 'voiced aspirates' was tenseness must be examined from the point of view of the phonetic nature of tenseness and its cross-linguistic distribution. The probability th:::t Proto-Indo­European possessed a series of tense voiced occlusives must be considered very low.

In the sense of the word tense as used by Chomsky and Halle, its main use is to distinguish tense and lax vowels, the former being produced with supraglottal muscular pressure, the latter without such pres~ure. Occlusives may also be tense, but in this case they are of necessity concomitantly voiceless: '•It is obvious that voicing can occur only if two conditions are met: the vocal cords must be in a position that will admit voicing, and there must be a flow of air through the glottis. When a stop is produced and the oral cavity is blocked while the vocal cords are in the appropriate position for voicing, pressure will build

up in the cavity and will rapidly ••• increase to the point where it is approximately equal to the subglottal pressure. This will halt the flow. of air through the glottis, thereby making further vocal vibrations impossible. Under these conditions there is only ane way in which the pressure buildup inside the vocal tract can be slowed down and voicing allowed fo take place during the

HOPPER: IE Occlusives- 147

dos11e phase of fre stop, that is by allowtng the weal tract to ellfland. If the walls of the tract are rigid as the result of muscular 1ensian, this expansion of the- cavity volume carmot take place, and ·therefore tense stops will nat shaw any wicing during the closure phase" (Chomsky and Halle, 1968: 325).

We may conclude that the distinguishing feature from a gene­tic point of view of fre 'voiced aspirates' cannot have been sup­raglottal tenseness, as in fact Chomsky and Halle point aut in their discussion of 'the aspirated. voiced steps of languages such

as Hindi' (326). We have seen that the 'voiced aspirates' cannot be said to

be aspirated in any I iteral sense of the word. Chomsky and Halle have assumed that they are prociJced with 'heightened subglottal pressure' (326), a feature which must be distinct from tenseness in that unlike tenseness it is not incompatible with -..oicing in occl usives. Yet Lcxlefoged reports that there is no clear experi· mental evidence for heightened. subglottal pressure in this type of obstruent, at least not insofar as such pressure is 'an inde­pendently controllable aspect of the speech event' (the condition placed by Chomsky and Halle on their features}; if there is an increase in subglottal pressure, 'it might he simply the result of variations in 1he state of the glottis, and not independently controllable in any way' {Ladefoged 1971: 96). Accordingly, any attempt to substitute a lenis/fortis distinction for a tense/lax one in the 'voiced aspirates' must also fail. In fact, Ladefoged's investigations suggest that phonological contrasts deriving from a difference in expiratory force are exceedingly rare among the world's languages, and there is no evidence at all that the phena· me non is draracteri sti c of lnda·E uropean.

2.5 Summary. The discussion so far has pointed fo two main conclusions about the PIE third series of stops: (1) that they were not phonetically 'voiced aspirates', and (2) that it is not appropriate to use the terms tense or fortis to distinguish them from the plain voiced steps. In the next section, I will discuss alternative possibilities concerning the phonetic and phonologi· cal nature of the PIE occlusive system.

3. It should be recalled that conclusions about the phonologi­cal or other systems of a proto-language ore drawn from two gen­eral sources: the comparative and internal data offered by the daughter· I ang uages, ond the constraints on I ang uoge structure 1r'hpo sed by uni verso! I ingui stic considerations. From the point

gloss" 7:2 :1973·,

of view of the. first of these sources, the positing of a voiced aspirate series made prima facie good sense. Unaspirated stops showed gaod sets of conespondences over the entire area of the language family, the main diver9ences being devoicing in Ger­manic and Armenian. In addition, there dearly existed a further class of o~clusives which showed up as plain voiced stops in some dialects (e.g. Slavic}, voiced fricatives in others (e.g. Germanic, in the earlier view), and aspirated stops in Greek and

Indo-Aryan. Comparative evidence supplied the phonological features

for this set of stops: voicing was indicated by most dialects, the main exceptions being Greek and Italic; stop articulation w~s also widespread, and languages which showed fricatives at all or some points of articulation, such as Italic and Germanic, also had stops as positional or dialectal variants. Aspiration was not so obvious, but the evidence of Greek and Sanskrit weighed heavily, because of their antiquity and prestige, even though they were in the numerical minority, and the fricot1ves in Latin and Germanic suggested a tensed or prolonged plasi ve consonant. Qualms about !he phonetic nature of 'voiced aspirated stops' could be eased by citing the supposed existence of such sounds in Sanskrit and the modem Indo-Aryan languages. Indeed we may assume that were it not for the direct evidence of these sounds in languages of India, including Sanskrit, the common etyma of words containing Greek kh, Germanic';; and Slavic g would hove been pes i ted on I y hesitantly and with m.Jch disagreement among scholars.

3.1 A New Hypothesis about the PIE Stops. Thus far I have questioned the usual description of some of the PIE obstruents on grounds of phonetic pr9bability, arguing that aspiration is incompatible with voicing: and that the features of Ienis versus fort1s, or lax versus tense, which linguists have sometimes posited as the phonological counterpart of aspiration in the voiced obstruents, are ether implau~ible or impossible when applied to the PIE 'voiced aspirates'. A second cogent argument against the system of consonants generally assumed for Proto­Indo-European is its typological improbability. Obstruent systems in which two voiced stops are opposed to a single voiceless stop at each point of articulation ore not unknown, but they are very rare, and of course such a system has not survived in any

of the Indo-European daughter-languages, although this fact

alone would not exclude it from the proto-language. Hockett

HOPPER: IE Occlus•ves- 149

(1955: 122-124) mentions only fwo languages with two voiced­one voiceless obstruent systems {Fiiian and the Mixtecan lan­guag~ Chatino). Of these, only rhe Chatino system, with voice­less, plain voiced, and prenasalized voiced stops, can be seri­ously considered, as the Fiiion obstruent system is hopelessly defective. The Choti no prenasal i zed stops ore suscep ti bl e of analysis as clusters of nasal plus stop, according to Hockett; o~e should, however, be cautious about analyzing away all sus­pected instances of unbalanced consonant sys terns. It would be most-unwise to regard the assumed obstruent system of Proto­Indo-European as impossible i.n the same sense as Dempwolff's reconstruction of Prato-Austronesian obstruents (referred to above 1) is. 'impossible'. It can, however, reasonably be regarded as imprabable, or at least as unusual enough to justify considering

alternative hypotheses.

3.1.1 The Phonetic Classification of Obstruents. Proceeding from Meillet's attribution to the voiced aspirates of 'a glottal articulation which we have no means of determining', I will begin by assuming that the voiced aspirates were stops with some kind

of voicing action, probably partial voicing or murmur. The exten­sion of this murmuring feature into the following vowel accounts for the 'aspirated' effect which observers notice in such sounds.

(closure) 1. Glottalic

2.

3. Creaky

4.

(full voicing) 5. Voiced

6.

7. Murmur

8.

(aperture) 9. Voiceless

Figure 3. Some States of the Glottis during Phonation (Adapted from Lade fag eel, 1 97l ).

glossa 7:2 !1973)

l

On the continuum between campi ete aperture and complete closure of the glottis (voicelessness and glottal stop respec­tively), it is possible to make a theoretically infinite number of distinctions. The recent survey by Ladefoged (1971) proposes a tentative nine-fold distinction, which, using Ladefoged's num­bering but ignoring several of the intermediate states, may be displayed as in Fig. 3. In addition, obstruents may be classified according to the point- again along a continuum -at which full voicing is initiated; this dimension is displayed by Ladefoged along the lines indicated in Fig. 4.

voicing

throughout

2 3

voicing immediately after

4 5

voicing considerably later

Figure 4. Timing of Voice On set (Ladefoged, 1971 ). 3.1.2 It is evident that, by combining these two tables, we will obtain o two-dimensional schema which will allow for con­siderable scope in positing obstruent phonemes for Proto-Indo­European without bringing in features such as tense, or positing two furly voiced phonemes against one voiceless. Figure 5 shows

Glottaiic

Creaky 3

Voiced 5 !d .. ' idh/?

Murmur 7

Voiceless9 . t '

Figure 5.

Voicing throughout

3

Voicing immediately after

5

Voicing delayed

Chart of laryngeal activity in obstruents (Ladefoged 1972),_ with platting of traditionally assumed Indo­European stops.

HOPPER: IE Occlustves- 151

the two tables combined, with a putative charting of the tradi­tionol~y assumed. PIE obstruenf system (with dental s as usual standing br eoch point of articulation). The anomalous nature of the term 'voiced aspirate' is highlighted by the attempt to place such a phoneme on the chart: it must itself be 'voiced', yet the initiation of voicing in the following phone must be 'delayed'!

If we assume that the obstruent system of Proto-Indo-Euro­pean was characteri ~ed by a tripl.e manner contrast, then the tcible in Fig. 5 offers several typologically plausible altemati ves to the traditionally assumed system. The optimum one, it seems

to me, waul d be a system containing two voi eel ess and one voiced stop, perhaps with a contrast of aspiration in the voiceless stops, or perhaps with one of the voiceless stops glottolized. For a number of reasons, the most probable alternative system to the one posited traditionally is, I believe, one in which the murmured stop is the only one marked for voicing, and the assumed plain voiced stop is glottalic, i.e. marked l•checkedl, i.e. the corres­pondences between the two systems areas in Fig. 6. The assump­tion that the third series was represented by murmured stops is o concession to the phonetic realization of these phonemes in the Indo-Aryan languages and in Greek, Latin, and other dialects where the evidence suggests a stop which was not of the plain voiced variety.

"Classical" System

d

dh

Proposed Revised System

t'

d

Figure 6

3.1.2.1 In an important article 'A reformulation of Grimm's Law', Emonds 11972) has suggested that an original system identical to the one proposed here except that the traditional 'votced stops' are canstciered platn lox voiceless stpps li.e. 'ti) and the traditional ·niceless stops are tense ( · th ·, etc.)

can account in o simp:e manner for developments in various Indo­European dialect areas. Thus the '-Oicing of the laxed stops is

common to a Central are<:x, and the appearance of voi eel es s stops in Germantc, A.rmeni'Jn, <:;n<:l 1oc-:ording to Emends) Htttite is

glossa 7·2 11973;

regarded as a relic area phenomenon. In all probability the views expressed in Emends' treatment and those presented here are not incompatible. I hove preferred to begin with the problem of typological plausibility, and believe that the evidence for a glottalic series is compelling enough to suggest at least o re­formulation of Emonds' work. In particular, Emends' solution requires that in at least one dialect area (Greek) a stage must be posited ir. which the 'traditional' stop series appears, namely 1t/, /d/, 1dh ', His claim that the traditional formulation cannot express in distinctive feature terms the generalization that s·uf­fixes may not begin with the 'voiced stops' is true, yet his own formulation ("IE suffixes beg1n only with tense stops" 115) seems strangely unmotivated; in the present treatment, suffixes may not begin with checked consonants - a situation true also

of Georgian (cf. 3.2.7. below). Future discussion of the whole question will, l hope, clarify the question whether the two view­

points can be reconciled, i.e. whether Emends' rules can be modified so as to accept proto-language forms containing glot­talics in place of lax voiceless stops.

3.1.3 The substitution of a glottalic series for the plain voiced stops has great advantages from a typological point of view, and can also be justified on grounds internal to the Indo­Europea-n language family, as I hope to sho"Y later in this paper.

Phonetical I y, glottclized sounds may be of several types. The term 'glottalic' or 'glottalized' has been used in at least two ways by linguists: (1) to denote the specific class of ejec­tives (sounds pronounced with pharyngeal air only and followed by a glottal release), an~ (2) as a cover term for laryngealized sounds and sounds chmocteri zed by upward or downward move­ment of the larynx (ejectives and implosives). l shall use the term in a somewhat narrower sense than the second of these, equating it with Ladefoged's feature of 'glottalic stricture'. I shall assume that loryngealized variants of ejectives developed into voiced stops over a large area of the Indo-European speaking

territory. !...adefoged's analysis of House describes a situation in which ejectives and laryngealized stops are never in contrast, and are subsumed as a natural class under the some feature of gl otto! i c stricture:

"[Housel contains a set of voiced stops, a set of voiceless

stops, and a set of glatto[i zed stops which includes the laryngealized sounds band d and the ejectivek'. The Iotter sound ••• can be scrid to have a glottal closure ·state l

HOPPER: IE Oeclust,es - 153

throughout the articulation and for a short penod after its ---·-release, before the beginning of regular voicing. It thus

appears on the chart very dose to .2 and g. At the systema­tic phonetic level all these sounds may be said to hove the maximum degree of glottal stricture and may be marked 1/, ~ausa having the possibilities /1 2 3/ for the feature at this level" (ladefoged, 1971, 21-22). Using Ladefaged's two-dimensional chart 1o plot the degree

of giottalic stricture beside the timing of voice onset, the pro­posed revision would give Proto-Indo-European a system of oc­clusives as in Fig. 7. In this Figure, dotted lines indicate the

Glottal ic

Geoky 3

Voiced 5

Murmured 7

Voiceless 9

-------------------- t' ·------------------

Voiced throughout

----------- d

·---------t--------------------3

Voiced

immediate! y after

5

Voicing

con si derabl y later

Figure 7. Hypothesized Proto-Into-European Stops. (Dotted lines represent allophonic and dialectal range of proposed phonemes)

range of allophonic and dialectal variation of each type. It must be remembered that the hypothesized range of these variants is much greater than would probably be found in a single language crea; this broadness is a reflection of the lock of refinement possible in specifying features of a reconstructed language.

3.2 The Indications for a Glottalized Series.

3.2.1 Typological Indications. Whereas obstruent systems with voiceless, pia1n V01ced and voiced aspirate (murmured) stops are rare or nonexistent in known languages, the type rep­resented by the pr<:Jposed re·Jisi'ln is both ccrmmon, and impor­tantly, has a wide geo<Jraphi-:'ll and cultural spread. It is repre-

glo~sa 7:2 n ?7;1

sented in 1he Old World by the Caucasian languages, in Africa by Hausa, and in the Americas by Quechua and a number of North American (·anguages. The Georgian system of stops, to give one example, is as follows:

Voiced b d 3 g

Voiceless p t c c k

Glotta!ic p' t' c' c' k' q'

3.2.2 Areal Implications. If, as seems probable, the fo·cal region of the Indo-European languages is to be sought in the area of the Black Sea and the Cis-Caucasian Plain, we might profitably look for structural similarities between Proto-Indo­European and languages assumed to have been spoken in that region. The work of the Georgi"an linguist Thomas Gamkrelidze (1966) points toward surprising parallels between Indo-European and Proto-Kartvelian (South Caucasian) in morpheme structure conditions. Further similarities ore noted by Schmidt (1967). These parallels. might, of course, be purely typological, but the strong possibility of earlier geographical ties between the two language groups must be reckoned with. Since glottalic conson­ants are a pervasive feature of all three Caucasian families, the suppositi_on of such sounds in Proto-Indo-European would add strength to Gamkrelidze's idea of an 'Indo-Caucasian' Sprachbund.

3.2.3 Proto-Indo-European ·b. The ranty or absence of the phoneme b' in Proto-Indo-European is well-known, and most of the handbooks comment on it. In the medial position, b occurs in some dialects as a sporadic voicing of p, as perhaps in Latin bibo 'I drink' Sanskrit pi~ati 'drinks'. Initially, examples of 'b are difficult 1o find; those usually cited ore often onomatopoeic or expressive, or else are restricted to o smaller number of dialects. The set of cognates often cited is Latin (de)-bilis 'lacking strength', Sanskrit ball" strong", Greek beltion 'better', etc., but other examples ore dubious.

From the point of view adopted here, the missing phoneme is the glottalic bilabial stop 'p' ·.It turns out that a gap in the labial series of a language with glottalized stops is a common situation. Greenberg, in his valuable paper "Some Generalizations concerning Glottalic Consonants, especially lmplosives' (1970) states: 'Preferences regarding point of articulation for glottalic obstruents are summarized in the followrng formula: injectives tend to have front articulation, ejectives to have bac.k articula­tion.' He goes on to observe:

HOPPER: IE Occlusives --155

•• A gap in 1he class of ejec1ives « the bilabial point of articulation is found in a number of world areas ••• Some Caucasian languages of· the North-East group, namely Avar, Andi, and Lak ••• ·have an extensive series of glot­talic. ejec.tives, which once more show a gap in 1he bilabial position" (127)

The phonetic reason for this seemingly arbitrary preference for back articulation in ejectives is apparent: the articulatory energy inv_alwd in producing such slops is greater the larger the air chamber between the glottis and the articulator. If the mi ss.ing PIE phoneme *b w~s in fact aligned with the glottalic series, then its absence is explicable on general typological and, even­tually, phonetic grounds.

3.2.4 The 'Marked' Sta1us of the Plain Voiced Stops. Meillet points out that the phonological role of the plain ·voiced stops was for less than that of the voiced aspirates or the voice! ess stops. He remarks:

"In historical times, especially in Indo-Iranian and Greek, the system of plain voiced stops is in contrast .·Nith thotof plain wiceless stops. But certain features indicate that, at least at a very early date, the system of plain voiced stops was not as essential in Indo-European as were those of the plain voiceless stops and the vaiced aspirate stops. Whereas *p is as essential as *t and *k, and *.bh as essential as *dh and *gh, *b is in Indo-European a'"rcrity, and betrays itself as of secondary origin in port" (Meillet, 1936, 84).

Hand-in-hand with !his observation go the statistical rarity of the

pi ain voiced stops in general,•and the constraints on the phonemic structure of roots. These constraints are discussed. in more de­tail below, but for the moment we will notice that the plain voiced stops are the only series which may not occur twice in the same root. It looks as if the 'plain wiced stops' represented. a marginal set of phonemes in Proto-Indo-European. Yet this s.ituation is surely anomalous, since the mast highly marked obstruents from the point of view of primary feature specifications were the -.oiced aspirates, which were marked(+) for·.both voic­ing and tenseness (aspiration, or whatever). We should somehow expect that in a ICJ'lguage with fully voiced, partially voiced, and unvoiced stops the primary phonological role waul d be be­tween the fully voiced and the unvoiced stops, i.e. t-d-s!, rather than t-s.!:d. This feeling is solely impressionistic, however, since

I know of no language having a double opposition of precisely

glossa 7:2 (1973"1

this lcincl. Assuming a glottalic series in place of the plain voiced

series, the marking distribution is a highly natural one. Accord­ing to Trubetskoy (1969), 'If ••• the correlation of recursion is one component of the correlation bundle, the 'mid' member of the gradational series is a (voiced or voiceless) Ienis with in­fraglottal expiration (t-d-t')' (152-153).

A glottalic series of consonants might be expected typo· logically to play a minimal role in suffixation, and this is cer­tainly true of the 'plain voiced stops' of Indo-European (Meillet 1936: 84). Languages in general display a preference for 'simple' phonemes in affixes, such as [n], [s], [t], etc. The generalization is difficult to formalize, but the truth of it is intuitive to the experienced linguist. In Georgian, voiced end voiceless conson­ants occur freely in affixes: [-eb], noun plural suffix; [g-] 2nd pers. sing, object; [-t] 1st/2nd pers.plur. subject(verbal affixes); etc. On the other hand, no inflectional affixes contain glottalics. The glottal ic consonants appear almost to be on the periphery of the consonant system (their frequent occurrence in nursery words and onomatopoeic words, in both Georgian and Indo-Euro­pean, is probably also symptomatic of this situation}.

The peripheral nature of the glottalics is supported also by an observation of Greenberg's concerning the widespread cross­lingui.stic tendency for glottalic consonants to lose their feature of glottalicity (1970: 134). This observed tenjency would account for the absence of a glottaiic series of obstruents in general in the recorded dialects of Indo-European.

3.2.5 Distinctive Features of PIE Stops. According to mod­ern phonological theor)', • phonological distinctive features are specified at some point (or perhaps interpreted) with the sign M or U (marked or unmarked), and these signs, if specified, are later converted into (.) or (--) by univ_ersal rules. The 'cost' of U F, as opposed toM F, (where F, is any feature) must be rec· koned in terms of the total grammatical system, and not merely the initial matrix, so that we must eventually take into account the number of times a feature will be mentioned, in the rules and the lexicon, before assuming that a specification M F, is a priori more cost! y than U F ,.

The assumption of plain and aspirated voiced stops in Proto­Indo-European, as has been pointed out, entails the assignment of the highest marking value to the voiced aspirates, i.e.:

HOPPER: IE Occ:lusiv~t$- 157

voiced

aspirated

t d dh

U M M

U U M

Accordingly, occurrences of /dhl in the rules and lexicon of Proto-Indo-European will be more costly than occurrences of /d/. Yet this situation is inconsistent with the phonological role of the 'voiced aspirates'. It could, of course, be argued that th~ unmcrlced value of aspiration in Proto-Indo-European was (+)but this would seem to be at variance with the condition on naturalness. A more natural system for Proto-Indo-European would be one in which only t'Ml M's appeared, e.g.:

F,

F2

(t) (d) (dh)

u u U M

M

u (where F 11 F 2 are manner features). This in fact is precisely the system which would have existed if one of the series .was glottal i c:

voiced

checked

t t' d

U U M

U M U

A further advantage which now accrues is that it is possible to refer lo the 'plain voiced' stop by a sing I e feature, and the voiceless stops and the murmured stops form a natural class of [-checked} stops.

There is one aspect of Indo-European phonology which is, I believe, considerably clearer if the new alignment is assumed: the constraints on the phonemic structure of the roots.

3.2.6 Root Structure Constraints in Inc/a-European. It is known that of the theoretically possible forms which a monosyllabic root of the type eve could take in Indo-European, se.veral types were excluded by systematic constraints on the cooccurrence of consonantal phonemes within the root. These restrictions were apparently as follows (Lehmann 1955: 17-18): A. The two consonants must differ in point of articulation regard­less of the manner feature.

!3. The root may not contai,n two plain voiced stops.

glossa 7·2 (1973i

C. The root may not contain both a plain voiceiess stop and a voiced as pi rate stop. Focussing on the constraints involving manner features ( B and C), we may sum up the restrictions as follows (assuming the vowel ie ', and using dentals for the initial stop and velars for the final stop):

Permitted Sequences

tek

teg

dek

degh

dheg

dhegh

Prohibited Sequences

~deg

*tegh

~dhek

In looking for a unified explanation for the constraints on manner features, we are struck by the apparent arbitrariness which they show. Why should sequences of voiceless sto.ps and sequences of voiced aspirate stops be allowed, but sequences of plain wiced stops prohibited? And what possible linguistic reason caul d there be for permitting types such as teg but ex­cluding types such os • tegh?

With regard 1o the exclusion of th; •tegh type -that with a voiced aspirate and a voiceless stop - we might claim that the constraint was one which forbade 1he co-occurrence of two ob­struents separated by more than one manner feature. Yet the principle does not seem to be assimilatory, for the opposite process operates ip the plain voiced stops. It appears as if the constraints B and C have 1o be considered irreconci I able: two unrelated means of restricting the possible root shapes.

Some of 1he arbitrariness of 1he root structure constraints is removed if the plain wiced stops are held 1o be glottalics. We have seen that 1he assumption of glottal ics brings about a realignment of the feature specifications such that the 'plain voiced' stops (i.e. the glottalics) now form a natural class in apposition to the wiceless and murmured stops as !-checked] versus 1--<:hecked[. We can exploit this ciass in making two succinct and I inguistically plausible statements of the con­straints:

(1) Each root contains at least one !-checked! obstruent;

HOPPER: IE Occlusives -15c;

(2) When both obstruents are [--checked}, they must agree in voicing.

The Proto.-lndo-European root structure constraints may now be represented as follows, where d, g are murmured slops (marked for voicing only at this level -;;f representation), and t', k' are glottal ic:

t,k g,g t',k' Permitted Prohibited

voiced - - - tek * t'ek'

checked - - + tek' *teg

t'eg *iek

.Q.ek'

t'ek

.s!e9

This restatement of the root structure constraints of Proto­Indo-European seems to have both phonetic and ci'Oss-linguistic phonological validity. Glo•talic stops, being articulated with supraglottal airstrean only, ore in a sense outside the voiced,' wiceless contrast, and are therefore neutral as to voice. Con­sequent! y they do not participate in the constraint which requires voicing agreement between stops in the sane root. The restric­~ion r11 has surprising parallels in three widely separated lan­guages having glottalized phonemes, Hausa (West Africa), Yucatec Mayan !Mexico), and Quechua (South America),

In Hausa, '. , • two heterorgon i c glottal i zed con son ants !including '? ') never coaccur in the same word' (Parsons 1970: 280). The constraint in Yucatec Mayan is identical. This lan­

guage hcs basic roots of the shape CVC, just as in Proto-lndo­Europe'ln before the addition of suffixes, and there ore both man­ner or.-:: ;:;oint of articulation constraints. One of these constraints is th~!: 'If beth consonants in the root are ejectives.,, then they rr.;.;st be i-lentical in every other respect' (Straight 1972: 59).

Accord:ng to Orr and Longacre (1968): " ••• in Quechuan forms f-:;u~d in our <:agnate sets, there is but one laryngealized stop :;r af+ric<Jte p~r 11ord. Thus where certain Aymara cognates 1n 'l Cul':<:hu-;,o•an set hove two loryngealized phonemes, the ~ o::-::::;n-:1 ; o• ;r.geoi • zed phcnem e wi II correspond to on unlaryngea­! •zed pf.c.nr:m~ ,n Prc.to-Ouechua, It therefore follows that all · ·c· -::n-:! · '(~1 ._:,-;,n-.meo; ha·1e not only the corresp?nding Proto-

g'ussa 7 2 lj'j/'j

Que chua laryngeal i zed reflexes but also reflex ~c. On the Quechuan

si.de, a dissimilation of laryngealized phonemes (to unlaryngea-1 i zed) affects the second I aryngeal i zed phoneme of any Quech­maran form. Thus PQA **t'ant'a gives Ay. t'an(a but PQ t'anta"

(549). In Proto-Indo-European there is no evidence that homorganic

or identical consonants escaped the stricture against two glot­

talics in the same root, since homorganic stops are, of course, prevented from occurring by the constraint A. The era ss-1 ingu is­

ti c tendency to avoid sequences of gl otto I i c phonemes reinforces

the hypothesis of the existence of these sounds in Proto-Indo­

European and their identification with the 'plain voiced stops'.'

3.3 I have argued that the positing of glottalic obstruents

in Proto-Indo-European not only gives the consonant system

greater typological plausibility, but is also capable of account­

ing for the root structure constraints. At this point it is necessary

to grapple with the difficult question whether we have not merely

pushed problematical aspects of Indo-European consonantism back into a 'Pre-Indo-European' period, or .....hether developments

in the Indo-European dialects can plausibly be traced direct! y to the obstruent system posited here. Natu roll y, we shall have gained little with respecT t~ typological plausibility if it is

necessary to admit some intermediate stage of Proto-Indo-Euro­pean identical in every respect to that traditionally posited. In this section, I hope to show that the consonant systems assumed to be present in the various sub-families of indo-European can either be derived immediately from the hypothesized basic system, or else are derivable through intermediate stages which have documented typological parallels.

3.3.1 The 'Sound Shifts' of Germanic and Armenian. It is well known that Germanic and Armenian share a remarkable systematic

change in their obstruent systems. In both languages, the Proto­

Indo-European 'plain voiced' stops became voiceless, and the 'voiced aspirates' became plain voiced stops. The original voice­less stops became fricatives in Germanic, and in AI'!Tlenian under­went a change to voiceless aspirates, with consequent further

changes according to the pi ace of arti cui ation. ~ From the point of v1ew of the original Indo-European ob­

struents adopted here, the changes which took place in Germanic and Armenian are seen as m1 nor phonetic differences. In parti cu­I or the supposed shift from voiced stops to vo1celess stops never

HOPPER: IE Occlusives- 161

occurred; instead, as Emonds has pointed out, Germanic and Armenian are to be viewed as 'rei i c areas' which were not affected

by the general Indo-European trend to realize glottalic stops as

fully voiced stops. It might now be objected that the Armenian system of obstru­

ents is remarkably similar to that of its long-time neighbor, Geor­

gian, which also has a contrast of voiced and voiceless stops and a full set of glottal ic stops. Clearly, the long symbiosis with Caucasian languages must be held responsible for the con­

tinued axi stence of Caucasian-! ike manner contrasts in Armenian; yet, as I have indicated, one should not ignore the possibility that Proto-Indo-European itself bore typological and areal simil­

larities 1o Caucasian languages. Armenian was spoken (or came to be spoken) in an area which was linguistically favorable to the retention of the original system.

3.3.2 Greek and Latin . • According to Meillet and Vendryes (1948: 69-74), the consonant system of Latin is traceable to a

system identical with that of an earlier stage of Greek, i.e.

p t k kW

b d g gW

ph th kh kWh

In this system, the Proto-Indo- European glottal i c stops appear as voiced stops, and the murmured stops ore represented by voiceless stops with aspiration. If these two changes were not

simultaneous, then the change from murmured stops to voiceless aspirates must have been first, since, otherwise, the manner contrasts would hove yielded t-d-!:J, the same system as the tra~ itionally posited one. I therefore assume a prior stage with three voiceless stops: 't/, 1 th/, and /t'l (plain, aspirated, and glot­

talic). There is ample typological precedent for this kind of

system, e.g. in Chippewyan, Chiricahuo Apache, and Yokuts

IHoijer, 1946). A second stage in which the glottalic stops be­came voiced would hove yielded the system underlying that of

Greek and Latin. In Latin, voiceless aspirated stops occur as 'f · (labials and dentals) or /h 1 (velars), sometimes with further

changes.

3.3.3 glottal i c

aspirate

Sansknt. In Sanskrit, as in Greek, Slavic, etc., the stops show up as ~.<>iced stops; and a distinct voiceless

series makes its appearance. The original murmured

giosso 7·2 (1973)

stops remain as the sounds traditionally spelled dh, bh, gh (with numerous changes internal to Sanskrit), After the genesis. of the voiceless aspirates, whatever their source, there may have been a stage in which three voiceless stops (/t 1, .'th ', 't'/) contrast­ed with a single voiced series (.f_g ). Unusual as it may seem,

there ore nonetheless languages known to have this configuration. Trager's analysis of Taos (New Mexico) (Hoijer 1946: 187) has the three voiceless stops and a voiced stop at labial, dental and velar points of articulation. Such a system, while highly marked, is thus not without documented parallels.

3.3.4 Adding 1o these examples language groups such as Celtic and Slavic, in which the glottolic stops and the munnured stops merged into a set of fully voiced stops, we may represent the trajectories of the Proto-Indo-European system of stops as in Figure 8.1t should be recalled that the purpose of this diagram, and of the discussion in this section, has been not to prove that the development of the Indo-European stops into the dialects was along the lines indicated, but simply to suggest that begin­ning with a Proto-Indo-European system of obstruents having one (partially) voiced, one unvoiced, and one glottolic stop at each

Greek Slavic Germanic Sanskrit Armenian

Stage I t t' d t t' si t t' d t t' .d. t t' 4 i I I I i I I I I I i I I I I I I

Stage II t t' th t t' d p t d th t t' .d. th t' d I I I ! \/ I I I

Stage Ill t d th t d t h t d Ji

Figure 8. Possible Trajectories of Indo-European Obstruents.

point of articulation, we do not need to assume any stage in which voiced sounds were in the numerical majority. Weak as this cloi m may appear to be, it is important to show that we can avoid positing typologically unprecedented systems at any point in the trajectory.

4. Conclusion. In this paper, I have discussed phonetic and phonological aspects of the Proto-Indo- European system of stops. I hove attempted to show that universal phonetic and systematic considerations I eod 1o the cone! usion that the sounds cl ossified as voiced tense or aspirated stops were probably partially voiced sounds (murmured stops), and that the supposed plain voiced

HOPPER: IE Occlusives- 163

stops were glottalized. This explanation accounts in a realistic way for the phonalogi col constraints on Indo- European root

structures.

NOTES

1-1 om grateful for comments and encouragement received from T. Gomkrelidze, E. Homp, R. King, Saul Levin, and E. Polome. Thanks are ol so due, less directly but nonetheless sincerely, to A. Kuipers, with whom I studied Georgian and Caucasian I on­guages at Texas in 1967. ln,response 1o my sending him a prior version of this paper, T. Gomkrelidze sent me a long and friendly letter describing the work of himself and V. Ivanov (December 1972) along closely similar lines to my own work. I find the close coincidence of our views, independently arrived at, immensely gratifying. It is my hope that the present article, besides pre­senting fr~h insights, will serve the function of supporting the Soviet (Russian and Georgian) scholars in their position, and helping make their views known in the English-speaking world. Any errors which remain are, of course, my own responsibil iry.

2-For a more detailed treatment of problems of this nature from a theoretical viewpoint, cf. Stockwell and Macaulay 1972.

3-Sequences of glottal ic consonants are found in Old Georgian, but under obviously definable conditions. Molitor's Glossar

1 1952) yields dear examples only in borrowed words such as p·inak'i 'key', p'et're 'Peter' (where Greek voiceless stops appear as glottal ics), and in tautosyllabli c clusters (t'q'avi 'skin, leather', etc.). In later stages of the language, borrowings are not so easily identified, and glottalic sequences are quite common. T. Gamkrelidze informs me in a personal communication r~at a constraint against heterorganic sequences of glottalics m~.ost be reconstructed for Proto-Kartvelion; a most important 'ypologi col and areal datum,

4-For the Germanic and Armenian sound shifts, cf. Meillet 1967: 116-124.

gl~s<c 7 2 ··1973;

REFERENCES Chomsky, Noam, and Morris Hal I e. 1968. The Sound Pattern of

English. New York, Harper and Row. Dempwolff, Otto. 1934. Vergleichende Lautlehre des Austronesis­

chen Wortschatzes, Band 1: I nduktiver Aufbau einer Indo nesi s· chen Ursprache. (Beiheft 15, Zeitschrift fur Eingeborene Sprach~n.)

Emonds, Joseph. 1972. 'A reformulation of Grimm's law.' In: M. Brame (ed.), Contributions to generative phonology. Austin, University of Texas Press, 1972.

Gamkrelidze, Thomas. 1966. 'A Typology of Common Kartvelian,' Lg. 42.1. 69-83.

Gamkrelidze, T.V., and Ivanov, V.V. Dec., 1972 'Lingvisticeskaja tipologija i rekonstrukcija sistemy indoevropejskix smycnyx,' Working Papers of the Conference on the Comparative-His­torical Grammar of the Indo-European Languages ( 12-14 December, 1972). Moscow, Akademija Nauk, 15-18.

Greenberg, Joseph. 1966. 'Some universals of grammar, with particular reference to the order of meaningful elements,' Universals of language, ed. by Joseph Greenberg. Cambridge, M.l. T. Press.

---~--· 1970. 'Some generalizations concerning glot­talic.consonants, especially implosives,' IJAL 36.123-145.

Hirt, Herman. 1927. lndogermonische Grqmmatik, I. Heidelberg, Carl Winter.

Hockett, Charles. 1955. A Manual of Phonology. (Memoir 11, IJAL) Baltimore, Indiana University Press.

Hoijer, Harry ed. 1946. Linguistic structures of native America. (Viking Fund publications in anthropology, 6.) New York, Viking Fund Inc •• •

Jakobsen, Roman. 1957. 'Typological studies and their contribu­tion to historical comparative linguistics,' Proceedings of the Eighth International Congress of Linguists (Oslo), 17-25.

Ladefoged, Peter. 1971. Preliminaries to Linguistic Phonetics. Chicago, University of Chicago Press.

Lehmann, Winfred P. 1955. Proto-Indo-European phonology. Aus· tin, University of Texas Press.

1972. 'Contemporary Linguistics and I ndo-Euro­pean studies,' PMLA 87.5, 976-993.

Meillet, Antoine. 1936. Introduction a I' etude comparative des langues indo-europeennes. Eighth Edition. (Reprinted as Alabama Linguistic and Philological Series, 3. University

HOPPER: IE Occ:lusives -165

of Alabama Press, 1964) Paris. 1967. The Indo-European dialects. English

translation, by Samuel N. Rosenberg, of Les .dialectes indo­europeens, 1908. (Alabama Linguistic and Philological Series, 15) University (Alabama), University of Alabama Press.

Meillet, Antoine, and J. Vendryes. 1968. Troite de grammaire comparee des langues classiques. 4th edition, by J. Vendryes.

• Paris, Champion. Mali tor, Ja seph. 1952. Altgeorgisches Glossar zu ausgewiihlten

Bibeltexten. Rome, Pontifi ci um lnstitutum Bi bl icum. (Man· umenta Biblica et Ecclesiastica, 6}.

Orr, Carolyn, and Robert Longacre. 1966. 'Proto-Quechumaran', Language 44,3: 528-555.

Parsons, F.W. 1970. 'Is Hausa really a Chadic language? Some problems of comparative phenology,' African Language Studies 11.272-288.

Peeters, C. 1971. 'A phonemic definition of 1-E bh, dh, gh', KZ 85.1.1-4

Prokosch, Eduard. 1938. A Comparative Germanic grammar. (William Dwight Whitney Linguistic Series.) Baltimore, Lin­guistic Society of America.

Schmidt, Karl Horst. 1967. 'Beitri:ige zu einer typologischver­gleichenden Gramniatik der indogermanischen und o;iidkau· kasischen Sprachen,' Miinchener Studien zur Sprachwissen­schaft 22.81-92.

Salta, G.R. 1963. Die armenische Sprache. (Handbuch der Orien­talistik, 1. Abteilung, 7. Band: Armenisch und Kaukasische Sprachen). Leiden/Koln, E.J. Brill.

Stockwell, Robert P., and R.K.S. Macaulay, eds. 1972. Linguistic change and generative lheory. (Indiana University studies in the history and theory of linguistics.) Bloomington, ln­di ana University Press.

Straight, H. Stephen. 1972. Yucatec Mayan pedolectology: seg­mental phonology. Dissertation, Department of Linguistics, University of Chicago.

Trubetzkoy, N.S. 1969. Principles of phonology. Translation, by Christiane A.M. Baltaxe, of GrundzUge der Phonologie (Prague, 1939). Berkeley and Los Angeles, University of California Press,

Whatmough, Joshua. 1937. 'Indo-European I abiovel or s, 'Festschrift

for Holger Pedersen, 45-56. (Received April 1973)

glossu 7:2 '1973)